Modelling Escherichia Coli Concentrations in the Tidal Scheldt River and Estuary

water research 45 (2011) 2724e2738

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Modelling Escherichia coli concentrations in the tidal Scheldt river and estuary

Anouk de Brauwere a,b,c,*, Benjamin de Brye b,c, Pierre Servais d, Julien Passerat d, Eric Deleersnijder b,e a Vrije Universiteit Brussel, Analytical and Environmental Chemistry, Pleinlaan 2, B-1050 Brussels, Belgium b Universite´ catholique de Louvain, Institute of Mechanics, Materials and Civil Engineering (IMMC), 4 Avenue G. Lemaıˆtre, B-1348 Louvain-la-Neuve, Belgium c Universite´ catholique de Louvain, Georges Lemaıˆtre Centre for Earth and Climate Research (TECLIM), 2 Chemin du Cyclotron, B-1348 Louvain-la-Neuve, Belgium d Universite´ Libre de Bruxelles, Ecologies des Syste`mes Aquatiques, Campus de la Plaine, CP 221, B-1050 Brussels, Belgium e Universite´ catholique de Louvain, Earth and Life Institute (ELI), Georges Lemaıˆtre Centre for Earth and Climate Research (TECLIM), 2 Chemin du Cyclotron, B-1348 Louvain-la-Neuve, Belgium article info abstract

Article history: Recent observations in the tidal Scheldt River and Estuary revealed a poor microbiological Received 27 July 2010 water quality and substantial variability of this quality which can hardly be assigned to Received in revised form a single factor. To assess the importance of tides, river discharge, point sources, upstream 31 January 2011 concentrations, mortality and settling a new model (SLIM-EC) was built. This model was Accepted 3 February 2011 first validated by comparison with the available field measurements of Escherichia coli Available online 13 February 2011 (E. coli, a common fecal bacterial indicator) concentrations. The model simulations agreed well with the observations, and in particular were able to reproduce the observed long- Keywords: term median concentrations and variability. Next, the model was used to perform sensi- Escherichia coli tivity runs in which one process/forcing was removed at a time. These simulations Fecal indicators revealed that the tide, upstream concentrations and the mortality process are the primary Microbiological water quality factors controlling the long-term median E. coli concentrations and the observed variability. Waste water The tide is crucial to explain the increased concentrations upstream of important inputs, Modelling as well as a generally increased variability. Remarkably, the wastewater treatment plants Scheldt discharging in the study domain do not seem to have a significant impact. This is due to Estuary a dilution effect, and to the fact that the concentrations coming from upstream (where Tidal rivers large cities are located) are high. Overall, the settling process as it is presently described in the model does not significantly affect the simulated E. coli concentrations. ª 2011 Elsevier Ltd. All rights reserved.

1. Introduction agriculture and animal farming, the Scheldt watershed (20,000 km2 from the North of France to the BelgianeDutch With its population density of more than 500 inhabitants per border, see Fig. 1) represents an extreme case of surface water km2, its active industrial development and its intensive and groundwater pollution (EEA, 2004). Improvement of water

* Corresponding author. Vrije Universiteit Brussel, Analytical and Environmental Chemistry, Pleinlaan 2, B-1050 Brussels, Belgium. Tel.: þ32 2 629 32 64; fax: þ32 2 629 32 74. E-mail address: [email protected] (B. de Brye). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.02.003 water research 45 (2011) 2724e2738 2725

Fig. 1 e Model domain and grid, showing the area of interest (Scheldt River and Estuary) covering only a small fraction, but containing a significant number of grid cells. (a) Complete mesh; (b) zoom on estuary and tidal rivers, also showing the connection between the 1D and 2D models, the different tributaries modelled as well as a few important locations. Important cities are encircled, sampling locations are indicated by coloured circles (blue: our monitorings, green: VMM stations, red: estuarine stations sampled during cruises). The same colours are used throughout the figures. Km indications refer to the longitudinal axis along the Scheldt used for visualising the simulations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) quality is however expected for 2015, owing to the ongoing climate.be/timothy). This must be achieved through the inte- implementation of the EU water framework directive (EU, 2000). gration of different existing and new mathematical models for Identification of pollutant sources, descriptionoftheir fate along describing the water flows and biogeochemical and microbial the Scheldt land-sea continuum and prediction of the evolution transformations for all aquatic compartments of the Scheldt of water quality in response to future implemented environ- land-sea continuum. The current study is to be situated in this mental policies and climate change e these are the objectives of broad framework, and more particularly focuses on the under- the Interuniversity Attraction Pole (IAP) TIMOTHY (www. standing of the microbiological water quality in the part of the 2726 water research 45 (2011) 2724e2738

Scheldt influenced by the tide. Recent field measurements through external sources. Regarding the tidal Scheldt River (Ouattara et al.,2011) have demonstrateda rather poormicrobial and Estuary, bacteria can enter through the upstream water quality in the Scheldt watershed concentrations above boundaries and tributaries. Obviously these inputs are highly the minimal water quality standards of the new EU Directive for variable. In addition, E. coli are brought into the domain by bathing water (EU, 2006). In addition, a large variability in the point sources of domestic waste water. Domestic wastewater measured concentrations was observed. Understanding these is released into the aquatic system after treatment in waste observations is the primary motivation for the current study. water treatment plants (WWTPs); the type of treatment The monitoring of microbiological water quality is based on greatly affects the concentration of fecal bacteria in the the concept of fecal bacterial indicators, whose abundance is released effluents (George et al., 2002; Servais et al., 2007b). related to the risk of pathogens being present (Havelaar et al., Wastewater discharges are expected to vary greatly on short 2001). Today, Escherichia coli (E. coli) is the more commonly time scales, especially during rain events. Finally, fecal used fecal bacterial indicator, as there was evidence from pollution can also be brought to surface waters through epidemiological studies (Kay et al., 2004) that its abundance is diffuse sources (surface runoff and soil leaching). In a recent a good indicator to predict the sanitary risks associated with study, (Ouattara et al., (2011) compared the respective waters (Edberg et al., 2000). contribution of point and non-point sources of fecal contam- E. coli concentrations measured in river waters often ination at the scale of the whole Scheldt watershed. They exhibit a variability which is so high that the concentrations concluded that point sources were largely predominant when are classically visualised on a log-scale. This variability is compared to non-point sources (around 30 times more for especially important in systems under tidal influence, as the E. coli at the scale of the Scheldt watershed). Predominance of part of the Scheldt studied here. Table 1 summarises the point sources was also demonstrated for the Seine watershed factors generally thought to affect E. coli concentrations and which is just south of the Scheldt one and is also highly variability in natural waters. However, it is often not clear urbanised (Garcia-Armisen and Servais, 2007; Servais et al., which factors are the main drivers explaining the mean 2007b). However, these results are based on catchment-scale concentrations and the concentration variability. calculations and diffuse sources can still have a significant Hydrological factors include the tide, river discharge and local impact on the E. coli concentrations, especially in small lateral runoff, which all influence the local transport, and rivers in rural areas. hence the local residence time, of the bacteria. These factors After their release in rivers, fecal bacteria abundance vary at different scales (and interact with each other); but it is decreases more or less rapidly. The disappearance of fecal clear that short term variations at the scale of the hour cannot bacteria in aquatic environments results from the combined be neglected. actions of various biological (grazing by protozoa, virus- Inputs of E. coli bacteria into the domain are also major induced cell lysis and autolysis) and physico-chemical factors controlling the E. coli concentrations in the system. conditions (stress due to osmotic shock when released in Indeed, it is generally assumed that fecal bacteria cannot grow seawater, nutrient depletion, exposition to sunlight and in natural water, and hence must be brought into the system temperature decrease) and also to possible settling to the sediments (Barcina et al., 1997; Rozen and Belkin, 2001). Unfortunately, it is difficult to identify the respective contribution of each of these factors to the decay rate at a given moment but it can be expected that their rate of disappear- Table 1 e Summary of factors affecting E. coli ance varies on timescales from hours to years. In models, the concentration in natural surface waters and the way these factors are represented in the model used in this decay of fecal bacteria is usually described by a first order study (SLIM-EC) for the Scheldt simulations. kinetics (Servais et al., 2007b). Factor affecting Representation in SLIM-EC From the above overview it is clear that many factors act on E. coli concentration the local E. coli concentrations, and most of them vary on short time scales. The goal of this study is to bring some insight into Hydrological factors the (relative) importance of these different factors in causing Tide 15 min resolution E. coli Upstream discharges Daily resolution the observed concentrations in the tidal Scheldt River Lateral runoff Parameterised (only in river part), and Estuary. The focus will be on understanding both the at daily resolution long-term median concentrations (varying in space) and the E. coli inputs local variability in concentration. For this purpose, the SLIM- Upstream concentrations Constant concentration EC model is set up which includes as many of these factors as (boundaries) possible (Table 1). This is the first E. coli model developed for Concentration entering Main tributaries explicitly in model the Scheldt tidal River, tributaries and Estuary, and the by tributaries WWTP point sources Constant discharge current paper presents the first realistic simulation results. Diffuse sources No As a number of factors can be included only approximately E. coli processes (due to a lack of information), it is not expected that concen- Mortality First order kinetic process, with trations can be predicted for a specific point and time. time-dependent coefficient (seasonal Furthermore, although the model is capable of simulating the variation linked to temperature) intra-tidal E. coli concentrations, the necessary high-resolu- Sedimentation First order process, coefficient vsed/H tion observations and boundary conditions are not available (with constant vsed) to evaluate the model performance at this scale. Rather, the water research 45 (2011) 2724e2738 2727

objective is to reconstruct the right median E. coli concentra- - by wind ﬁelds at 10 m above the sea level. These ﬁelds are 4 tions (taken over time periods of the order of one day to a year) times daily NCEP Reanalysis data provided by the NOAA/ and concentration variability, both in time and in space. The OAR/ESRL PSD; ability of the model to achieve this goal is assessed by - at the upstream river boundaries, the mouths of the Seine, comparison with the available data. Thames, Rhine/Meuse, the Bath Canal, Ghent-Terneuzen Canal and the Antwerp Harbour locks: by discharges inter- polated from daily measurements. 2. Model description

The model used in this study is a version of the Second 2.3. E. coli module generation Louvain-la-Neuve Ice-ocean Model (SLIM: www. climate.be/slim). As its name indicates, this model originally SLIM-EC combines the hydrodynamic SLIM with a module focuses on the physical processes in the aquatic environment, describing the dynamics of E. coli in the aquatic system. In this and does so by solving the governing equations using the module the bacteria are modelled as a single type of reactive finite element method on unstructured meshes (“second tracer, i.e. once they enter the model domain (through generation”). Unstructured grids offer the possibility of a more external sources), they are transported by the hydrodynamics accurate representation of coastlines and grid sizes varying in and their concentration is affected by E. coli-specific processes. space (and time) e without having to increase the total In the 2D part of the model domain, the depth-averaged number of discrete unknowns. A validated SLIM version for concentration C of E. coli is described by the following advec- the hydrodynamics in the Scheldt (de Brye et al., 2010)is tion-diffusion-reaction equation: combined with a simple reactive tracer module for the simu- v lation of E. coli concentrations, forming SLIM-EC. Table 1 ð ÞþV ð u Þ¼V ð V Þþ v HC H C KH C HR (1) summarises the main processes and inputs and at which t temporal resolution they are represented by the model. where t is the time, V the horizontal del operator, H the water depth, u the depth-averaged velocity vector, K the diffusivity 2.1. Model domain and mesh coefficient and R the reaction term (which will be described in more detail below). As the mesh size varies greatly over the The computational domain (see Fig. 1) is identical to that used computational domain, it is essential to that the horizontal by de Brye et al. (2010): although the focus is on the Scheldt diffusivity varies with the mesh size. In this study the diffu- Estuary (indicated by the rectangle in Fig. 1a and shown in the sivity coefficient K depends on the mesh size D according to zoom of Fig. 1b), the domain is extended both upstream and a relation inspired by Okubo (1971): K ¼ aD1.15, with downstream. Upstream the domain reaches as far as the tidal a ¼ 0.03 m 0.85s 1. influence is significant, covering a riverine network of the In the 1D part of the model the following advection-diffu- Scheldt and its tributaries. So, although the Scheldt is the sion-reaction equation is solved for the section-averaged main focus of this study, all main (tidally influenced) concentration C of E. coli: tributaries are also modelled explicitly. This riverine part of the model is 1D (averaged over the cross section), while the v v v v ðSCÞþ ðSuCÞ¼ KS þ SR (2) estuary and the downstream extension covering the whole vt vx vx vx North-Western European continental shelf are modelled by where S is the section of the river and u the section-averaged 2D, depth-averaged equations. velocity. The variable x represents the along-river distance. Fig. 1 also shows the unstructured mesh used, constructed The processes affecting E. coli concentration in the water by Gmsh (Geuzaine and Remacle, 2009; Lambrechts et al., column that are considered in the SLIM-EC model are 2008), which is made up of approximately 21,000 triangles (in mortality and sedimentation. The approach used to model the 2D part) and 400 line segments (in the 1D part). In the these processes is similar to that of Servais et al. (2007a, b) to current study a mesh was used with triangle sizes covering model the dynamics of fecal coliforms in the rivers of the several orders of magnitude (the ratio of the size of the largest Seine drainage network. Both mortality and settling are triangle to the smallest exceeds 1000, the smallest with modelled by first order (type) reaction terms: a characteristic length of w60 m are in the Scheldt Estuary). For a more detailed discussion of the computational domain and vsed R ¼kmortC C (3) construction of the mesh, please refer to de Brye et al. (2010). H

The sedimentation velocity vsed is assumed to be constant and 2.2. Hydrodynamics equal to 5.56 10 6 ms 1. This value is based on experiments conducted to study the fecal bacteria settling rate in rivers A detailed presentation and validation of the hydrodynamical from the Scheldt and Seine watersheds (Garcia-Armisen and model SLIM can be found in de Brye et al. (2010). We only Servais, 2008). Note that this representation of the disap- repeat here the aspects determining its temporal resolution. pearance rate by sedimentation is a parameterisation for The model has a time step of 15 min. It is forced depth-averaged models, implying that the water column is well-mixed. In practice, this assumption may not be entirely - at the shelf break: by elevation and velocity harmonics of valid, but it has been shown that the error made remains the global ocean tidal model TPXO7.1; relatively small (de Brauwere and Deleersnijder, 2010). 2728 water research 45 (2011) 2724e2738

The mortality rate varies with temperature following the average volume treated in the WWTP (which depends on a sigmoid relation (Servais et al., 2007a, 2007b): the number of inhabitants-equivalents connected to the ðT25Þ2 WWTP) multiplied by an E. coli concentration depending on the exp À 400Á kmortðTÞ¼k20 (4) treatment type applied in the WWTP (George et al., 2002; exp 25 400 Servais et al., 2007b). The E. coli concentrations considered in ¼ 5 1 5 1 with T representing temperature in C and k20 1.25 10 s . the treated effluents was 2.8 10 E. coli (100 ml) when a the We do not have high-frequency high-resolution temperature primary treatment followed by an activated sludge process was 5 1 measurements in the Scheldt. But using the temperature applied, 1.7 10 E. coli (100 ml) when the N removal treat- measurements made at the monthly intervals during ment (nitrification þ denitrification) was added to an activated 5 1 2007e2008 at several locations, we fitted a sine through these sludge process and 1.1 10 E. coli (100 ml) when the treat- points in order to get the average seasonal temperature in the ment included activated sludge followed by N and P removal; whole domain as a function of time (Fig. 2). Using this relation, these values result from measurements performed in treated we can now approximate the temperature at any time during effluents of various WWTPs located in the Scheldt watershed. the simulations. Substituting this in equation (4),weeffectively After this procedure, the E. coli discharges in the model domain 6 1 8 1 get a mortality rate varying seasonally. The value of the by the WWTPs ranged from 8 10 s to 8 10 s . mortality rate was similar to the one used byServais et al. (2007a, b) for modelling the dynamics of fecal bacteria in the Seine 2.4.2. Open boundary concentrations watershed. We verified in batch experiments (data not shown) The concentration of E. coli entering the domain through the that the mortality rates in the large rivers of the Scheldt water- open boundaries must also be assigned (see Fig. 1 for location shed were not significantly different from those estimated for and Table 2 for values). The concentration at the shelf break the large rivers of the Seine watershed. In this model, to the was assumed to be zero, as well as the concentrations entering “base mortality” no additional mortality term was added related the estuary laterally (the Bath and Terneuzen Canals, and to solar effects, as is done in some other studies (Liu et al., 2006; water coming from the Antwerp harbour locks). The assump- Thupaki et al., 2010). The main reason for this is that in the tion for the shelf break seems undisputable, due to its large modelled domain water is quite turbid (from 20 mg/l of sus- distance from land. The concentrations in the canals were not pendedmatter to more than 1 g/l in the maximumturbidity zone measured but estuarine observations indicate that their effect of the estuary), resulting in a low light penetration and thus is very limited (see below). The effect of the harbour was a limited impact of solar irradiation on fecal bacteria. neglected based on specific measurements made inside and outside the locks, which were quasi-identical (unpublished data). Furthermore the harbour authorities estimated the 2.4. Input of E. coli into the system average residence time in the harbour to be of the order of several months, suggesting that bacteria entering the harbour 2.4.1. Input by WWTPs are probably long dead before they could reach the locks. As Ouattara et al., (2011) showed that E. coli enter the Scheldt The only boundaries through which a significant amount mostly through point sources (cf. Introduction), WWTP outlets of bacteria enters the domain are the upstream river bound- are the only sources included in the model (see also Table 1). aries. These boundary concentrations are based on field WWTP data are compiled from information provided by the measurements taken at the boundary locations (unpublished Vlaamse Milieumaatschappij (Flemish Environmental Agency, data). If only one measurement is available, this value was VMM), Rijkswaterstaat Zeeland and Waterschap Zeeuwse considered, otherwise the median value of all measurements Eilanden for the whole (tidal) basin. Data processing steps available at that point was used. The data did not allow to involved the localisation of the WWTP outlet, the actual impose boundary concentrations varying in time e although discharge point in the model domain, and the distance we did investigate whether the measured concentrations between these two points. The number of E. coli discharged by correlated with discharge, but no significant relation was a WWTP per second was approximated to be proportional to revealed (Ouattara et al., 2011).

temperature (°C) 10

4 27.03.2007 28.05.2007 25.07.2007 24.09.2007 06.11.2007 04.02.2008 17.03.2008 14.05.2008 date

Fig. 2 e Fitted sine (black line) through temperature measurements made at several locations in the Scheldt and its tributaries (dots). water research 45 (2011) 2724e2738 2729

(1) Simulated median and range (over the period of the our Table 2 e E. coli concentrations imposed at the model boundaries in SLIM-EC. monthly monitoring) of E. coli concentrations along the Scheldt axis (Fig. 3) and along the Rupel-Nete-Grote Nete Boundary concentrations in E. coli (100 ml) 1 axis (Fig. 4) are compared to the median and range of Durme 2600 measured values. This enables an assessment of the Scheldt upper branch 10000 simulated median and variability, and its variation in space. Scheldt lower branch 15000 (2) Simulated and measured timeseries at a given point in Kleine Nete 1900 space (two locations in the Scheldt River, Fig. 6). This Grote Nete 1500 Dender 700 comparison more clearly visualises the simulated and Dijle 3400 measured long-term variability in time. Zenne 400000 (3) Simulated and measured concentrations on two speciﬁc Shelf break, rivers discharging in North Sea 0 days, at a number of speciﬁc estuarine stations (sampled and canals discharging in estuary during two cruises, Fig. 5). This comparison focuses on the estuarine part; it visualises the short term model variability, but only point-wise comparisons with the obser- 3. Validation measurements vations are possible.

The E. coli concentrations calculated by the model were Fig. 3a shows that the model is able to reproduce the compared to field measurements made in the study domain in measured median concentrations and concentration range in order to validate the model. The modelling period was chosen the tidal Scheldt River (1D model). The median values corre- such that it covers the measurements made in the scope of the spond very well to the observed medians (Table 3). The differ- IAP TIMOTHY project, i.e. February 2007eJune 2008. Two types ence certainly falls within the measurement precision of of sampling campaigns were conducted during this period: approximately 25% (cf. section 3). On the other hand, it appears that the model finds a larger range of concentrations than From 26 March 2007 to 13 June 2008, monthly samples were those measured (when expressed as interquartile range, cf. taken at several monitoring stations in the Scheldt water- Table 3). This is probably due to the fact that the model covers shed. This gives monthly timeseries at several locations, but a much wider range of hydrological regimes than the monthly also enables to assess the long-term variability. measurements. Indeed, the modelled range is primarily In February 2007 and 2008, two one day cruises along the a reflection of extreme events occurring during the simulation saline estuary were conducted. This resulted in two longi- period. It is not surprising that these brief extreme conditions tudinal estuarine profiles. are not captured by a monthly point sample. Furthermore, it was attempted to carry out the monitoring samplings The results of the latter monitoring survey are fully approximately at low water, but due to logistic constraints this described in Ouattara et al., (2011). E. coli concentrations were is not exactly the case for all stations. This could be an addi- estimated by a plate count method using Chromocult Coli- tional factor lowering the observed range. form agar medium. By performing replicates, the coefficient of According to Fig. 3 the WWTPs have little effect on the variation (CV) of the plate counts on specific media used in concentrations, while the tributaries and the water from this study was estimated to be 25%. This value of CV is usual upstream have a more significant influence. This is especially for this type of bacterial enumeration (Prats et al., 2008). true for the water coming from the Rupel, as this river also In addition, a second set of data was used: measurements carries water coming from the Zenne crossing the city of of fecal coliforms made by the VMM at one station in the Brussels (cf. Fig. 1). Fig. 4 shows the simulation results for the Scheldt River (Zele) and three locations in the estuary very Rupel, including the measurements made during the monthly close to each other (around Doel). The fecal coliform concen- monitoring, clearly illustrating the huge concentrations trations were converted into E. coli concentrations by multi- entering through the Dyle/Zenne. Ouattara et al., (2011) plying the fecal coliform data by 0.77; this value is the average reported on the Zenne water quality in more detail, noting ratio between E. coli and fecal coliforms numbers measured in that the section downstream of Brussels is heavily contami- river water samples (Garcia-Armisen et al., 2007). The VMM nated with E. coli abundances comparable to those usually measurements span different periods, ranging from 2000 to measured in treated waste waters. 2008, and hence do not exactly correspond to the modelled The effect of the tide is also clearly visible in Figs. 3 and 4, period. Therefore, these measurements should be regarded as high concentrations are also transported upstream of the with some caution. input point (e.g. when the Dyle/Zenne join the Rupel in Fig. 4, or when the Rupel joins the Scheldt in Fig. 5). Indeed, the tides periodically push water up the rivers, thus counteracting the 4. Results and discussion “normal”, downstream directed, river flow. Without tides, the high concentrations would primarily be transported down- 4.1. Reference simulation stream. This important feature could only be captured by a model resolving tides, and suggests that the tidal process The simulations are compared to the available observations in may indeed be an important factor explaining the observed three different ways, enabling model validation from different concentrations and/or variability. In particular, it seems that perspectives: the concentrations measured at Temse (Fig. 3a) are highly 2730 water research 45 (2011) 2724e2738

4 x 10 4.5

Zele a ← Berlare Temse Bornem

4 Wetteren ← ← ← ← Overschelde Destelbergen Sint−Amands Dendermonde ← ← ←

3.5 ← Lede & Wichelen ← Aartselaar ← −1 3

2.5

1.5 E. coli (100 ml)

1 Uitbergen Temse 0.5 Zele 0 0 southern ↑ 10 20 30 Dender ↑ 40 50 Durme ↑ 60 Rupel ↑ 70 Scheldt branch distance from Ghent (km) − 1D model

3000 b Burcht SLIM−EC interquartile band ←

−1 2500

SLIM−EC min − max values Walcheren SLIM−EC median value ← 2000 Bath & Waarde Antwerpen−Zuid ←

← measurements Willem Annapolder

1500 ←

1000 E. coli (100 ml) 500 Doel

0 80 90 100 110 120 130 140 150 distance from Ghent (km) − 2D model

Fig. 3 e E. coli concentration profile along the Scheldt, from Ghent (km 0, cf. Fig. 1) to the mouth. (a) Results from 1D model. (b) Results from the 2D model. Red vertical lines indicate the location of the WWTPs, blue vertical lines the location of tributaries joining the Scheldt. Only the simulation results covering the our monthly monitoring period are considered. The simulations are summarised as their median value at every position (black line), the interquartile range (grey band) and the min-max range (dotted lines). The available measurements are shown as dots: cyan dots referring to our monthly monitoring, and green dots referring to VMM measurements (only the bigger dots represent samples taken during the simulation period), squares indicate the median value of the measurements at each location. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) influenced by the Rupel although Temse is situated upstream suspended matter. For instance, explicitly modelling resus- of the Rupel connection. The importance of the tide will be pension and longer survival times for E. coli bacteria attached to further discussed in section 4.2. sediment particles (Craig et al., 2004; Davies and Bavor, 2000; In the estuary, the major feature is a steep decrease in simu- Davies et al., 1995) may indeed increase the modelled concen- lated concentrations (Fig. 3b).Thisdecreaseiscoincidentwiththe trations in the MTZ. A second possible explanation for the model maximum turbidity zone (MTZ) in the Scheldt, which is reported underestimation is missing sources. WWTPs are included in the approximately between km 60 and 100, or between salinity values model, but not the possible pollution effect of canals, or of diffuse 2and10(Baeyens et al., 1998; Chen et al., 2005; Muylaert and sources (most of the estuary lies in a rural area). Sabbe, 1999). Measured timeseries in the estuary are scarce. The In Fig. 6 the model results are visualised as timeseries at two only timeseries in the estuary available to us are those performed monitoring stations in the Scheldt. These figures visualise more by VMM. As discussed in section 3, these measurements are to be explicitly the temporal variability in the observations and interpreted with care, but it appears that the model underesti- simulations. It can be seen that the model is not able to repro- mates the concentrations in this part of the estuary, or at least duce the observations exactly, i.e. the model is not accurate for cannot reproduce some of the higher values measured. The predictions of the exact concentration at a given time and model performance in the estuary is further assessed in Fig. 5, location. However, the median value and range are satisfactorily comparingmeasurementsmadeduringtwoestuarinecruises modelled, especially when comparing with the generally with model outputs from the same days. These results suggest reported performances of microbial quality models described in that the model predicts the correct concentrations in the begin- the literature, where one is generally satisfied with model ning and at the mouth of the estuary, but simulates too fast simulations falling within half a log unit of the observations a decrease between these two extremes. Again, the concentration (Collins and Rutherford, 2004; Garcia-Armisen et al., 2006; decrease occurs in the MTZ. Therefore, the poor model perfor- Sanders et al., 2005). The modelled variability has a different mance in this part of the Scheldt is probably related to the fact nature at the two locations: in Temse (Fig. 6a) a large portion of that the E. coli dynamics are modelled as independent of the variability is due to the tide (compare raw outputs with water research 45 (2011) 2724e2738 2731

5 x 10 upstream than Temse. It is also in agreement with Ouattara

et al., (2011), who identiﬁed a positive correlation between E.

2.5 Boom coli concentrations and discharge at Uitbergen, while there was ← no signiﬁcant correlation at Temse. The tidal inﬂuence in Temse was already suggested when inspecting Fig. 3a, and is related to 2 the Rupel joining the Scheldt downstream of Temse. The high

−1 E. coli concentrations carried by the Rupel are pushed upstream (to Temse) at a tidal frequency, explaining the important tidal 1.5 fingerprint in the timeseries at this location. Conversely, at Uitbergen, there is no important source in the vicinity which could cause a similar tidal influence. 1 E. coli (100 ml) In this section, the model results of the reference simulation were assessed and generally a good agreement is found Boom 0.5 for the median concentration and its variability. This validation is not trivial as the model parameters (for mortality and Duffel settling) and inputs (WWTPs and boundary concentrations) 0 were not tuned, but directly taken from field measurements or 0 5 10 15 20 25 30 35 40 ↑ Kleine Nete ↑ Dijle/Zenne external studies. The (potential) influence of tide, river distance from upstream Grote Nete boundary (km) discharge, WWTP inputs and upstream concentrations have Fig. 4 e E. coli concentration profile along the Rupel-Nete- been briefly discussed. The importance of these factors will be Grote Nete axis (cf. Fig. 1). Km 0 refers to the upstream further investigated in the next section. boundary of the model in the Grote Nete. For legend refer to Fig. 3. 4.2. Impact of different processes on E. coli concentrations tidally averaged concentrations), while in Uitbergen (Fig. 6b) One of the objectives of this study is to better understand the most of the variability seems to occur at longer timescales and is importance of the different factors affecting the long-term probably more related to the hydrological regime. This agrees median E. coli concentration and its variability in the Scheldt with what we could expect as Uitbergen is located more River and Estuary. Starting from the reference simulation

8000

S20 a b 1400 7000

1200 6000 S18

1000 5000 S22 −1 −1

800 4000

S22 600 E. coli (100 ml) E. coli (100 ml) 3000

400 2000

S20 1000 200 S18

0 0 0 5 10 15 20 25 30 0 5 10 15 20 25 30 salinity S22 S18 salinity S15 S12 S09 S07 S04 S03 150 S20 S15 S12 S09 S07 S04 S01 150 S01

Fig. 5 e Estuarine proﬁles of E. coli concentrations on two speciﬁc days of longitudinal cruises in the estuary: (a) 14 February 2007, (b) 12 February 2008. Black dots represent the simulated E. coli concentrations at the same location as the cruise stations during the whole cruise day. The larger black circle shows the simulated value approximately at the time of sampling. The crosses represent the measurements. Station names are also added to facilitate localisation of the stations (see Fig. 1, station 150 is a sea station outside the mouth of the Scheldt). 2732 water research 45 (2011) 2724e2738

Fig. 6 e E. coli concentration timeseries at two locations in the Scheldt River (see Figs. 1 and 3 for location): (a) Temse, (b) Uitbergen. Simulation period covers our monthly monitoring. Black line shows the model output, grey line the tidal moving average of these outputs. Dots represent ﬁeld measurements made during this monitoring.

presented in the previous section, we removed, one by one, First inspecting what happens at the two monitoring the major processes (cf. Table 1). Table 3 summarises the stations Temse and Uitbergen (Table 3), it is seen that the results of these different simulations. change is largest at Temse. Indeed, both median concentration and variability (interquartile range) are significantly reduced. 4.2.1. Tide and upstream discharge Surprisingly, the median concentration at Uitbergen increases, To assess the role of the tides, a simulation was run with the while the variability remains equal. This confirms the tides removed from the hydrodynamics, while all other forc- hypothesis formulated when discussing Fig. 6 that Temse is ings and processes are kept identical. much more influenced by the tide, because it is the tide that allows water mass to flow from downstream to upstream and thus brings the high Rupel concentrations upstream. When the tide is switched off, the Rupel concentration cannot reach as Table 3 e Comparison of observed and simulated median and interquartile concentrations all expressed as E. coli far upstream anymore (Fig. 7). Fig. 8a shows the simulated L (100 ml) 1. The comparison is done at two monitoring timeseries at Temse, showing the reduced concentrations and locations, where samples were taken at approximately variability. The remaining variability is related to the upstream monthly intervals from 26 March 2007 to 13 June 2008. discharge (average daily discharges are prescribed). Fig. 8c The simulations cover the same period, but all model shows the daily water discharge at Melle (see Fig. 1 for location) outputs (at 15 min intervals) are used to compute the and there is indeed a clear similarity with the concentration statistics. timeseries at Temse. High concentrations at Temse generally Temse Uitbergen coincide with high discharge periods. Median Interquartile Median Interquartile The concentrations at Uitbergen are overallless influenced by range range the tide. Therefore, it is no surprise that the simulated concen-

Observations 1400 1200 3500 3700 tration timeseries at Uitbergen (without tide, Fig. 8b) also exhibits Simulations a clear similarity with the discharge timeseries, although the Reference 1500 3000 3600 4700 concentrations seem to be less “sensitive” to high discharges No tide 80 280 5500 4700 than was the case at Temse. This suggests that the two coun- No upstream 110 51 300 120 teracting effects of high discharge e reduced transit time conc. (increasing E. coli concentrations downstream) and increased No WWTPs 1400 3000 3300 4900 e ¼ dilution (decreasing concentrations) are balanced differently kmort 0 16000 20000 10000 2900 at these two locations. But the overall result at both locations is vsed ¼ 0 1900 3800 4700 5100 an increase of the E. coli concentrations with discharge. water research 45 (2011) 2724e2738 2733

4 x 10

Zele a ← Temse Berlare

5 Bornem Wetteren ← ← ← ← Overschelde Destelbergen Sint−Amands Dendermonde ← ← ← ← Lede & Wichelen

4 ← −1

3 Aartselaar ← 2 E. coli (100 ml)

1 Uitbergen Temse Zele

0 0 southern ↑ 10 20 30 Dender ↑ 40 50 Durme ↑ 60 Rupel ↑ 70 Scheldt branch distance from Ghent (km) − 1D model

3000 SLIM−EC interquartile band b Burcht

← SLIM−EC min − max values

−1 2500 SLIM−EC median value Walcheren ←

2000 Bath & Waarde measurements Antwerpen−Zuid ← ← Willem Annapolder

1500 ←

1000 E. coli (100 ml) 500 Doel

0 80 90 100 110 120 130 140 150 distance from Ghent (km) − 2D model

Fig. 7 e E. coli concentration proﬁle along the Scheldt (cf. Fig. 3 for legend). Model results refer to simulation without tide.

Further inspecting the simulations at Uitbergen without reveals a complex role of the tides: they can locally either tides, it is remarkable that the median simulated concentra- increase or decrease the median concentration and, surpris- tion is increased, while the interquartile range remains ingly, the same holds for the variability. Indeed, in the central unchanged. When comparing Fig. 8b with Fig. 6b, it appears part (between km 22 and 50) the tides effectively reduce the that switching off the tide induces two main changes: observed variability in E. coli concentrations. Further downstream (from km 50 to the Rupel) the tides hugely increase (i) the short term variability due to the tidal effect vanishes, both the variability and the median concentrations, until as expected. Because this variability has a smaller almost 100% of their value is due to the tides. This is the amplitude than the long-term variations, this barely upstream Rupel influence zone, as discussed for the sampling influences the overall interquartile range. station Temse. Upstream of km 50 the median concentrations (ii) the minimal concentrations are higher (although the are lowered by the tides (cf. discussion for Uitbergen), and this maximal concentrations remain quasi-identical). Indeed, reduction is higher than 50% for a significant section of the in the simulation with tides, the concentrations drop to river. lower, almost-zero values. As for Uitbergen no major In order to better understand why the tides reduce median sources lie downstream, during rising tide, waters with and interquartile range in the central part of the river, we lower E. coli concentrations are brought upstream to Uit- performed an additional model test. A narrow patch of tracer bergen, effectively reducing the concentration at Uitber- was initialised at Uitbergen on 1/2/2007 at 0:00 and followed gen. It is remarkable that the concentrations remain at during 10 days e once transported by the “full” hydrody- these low levels for significantly longer periods than namics (tides þ river flow), and once with only the river flow. a tidal cycle. Therefore, these low values cannot (only) For simplicity, all other sources and decay reactions were result from the periodic tidal current upstream. Rather, it removed (passive tracer). Fig. 10 shows the results of these seems that the tidal oscillation has a mixing effect acting two simulations. It is seen that, in addition to moving the on longer timescales, especially during periods of low patch up and down the river, the tides increase the width of discharge, when there is less counteraction from the river the patch and accordingly reduce the maximal concentration. flow. This suggests that the tides indeed have an increased “mixing” effect, smoothing the patch more efficiently than without These results clearly demonstrate that the concentrations tidal action, which is compatible with the observed lower at both monitoring locations are influenced by the tides, but in median concentrations and variability in this section of the a different manner. In order to get a more detailed picture of river. the spatially varying effect of the tides on median concen- In the estuary, the picture does not change so much by tration and variability, we visualised the differences between removing the tidal effect (Fig. 7b). Without tides, the high Fig. 3 (with tides) and Fig. 7 (without tides) in Fig. 9. This figure Rupel concentrations propagate less far downstream. Only 2734 water research 45 (2011) 2724e2738

Fig. 8 e E. coli concentration timeseries at (a) Temse and (b) Uitbergen (cf. Fig. 1 for location). Model results refer to simulation without tide. (c) Measured daily discharge at Melle.

the residual current drives the concentrations downstream, effect to decrease the concentrations by bringing downstream resulting in slightly higher concentrations close to the Rupel water which contains lower concentrations of E. coli. Finally, by and a faster decrease to quasi-zero values. removing the tidal forcing it was also clearly seen that both at In conclusion, the tide appears to have a significant influ- Temse and Uitbergen the modelled E. coli concentrations ence on the E. coli concentrations (median and range) e but the correlate positively with upstream discharge, although their effect is different depending on the location. Overall, the tide response is different. Clearly, the impact of the tides on the E. has the effect to enlarge the influence radius of a source (or coli concentrations is crucial but very complex, implying that tributary) by pushing water upstream and further downstream “tidal corrections” in models which would not explicitly than if there were no tides. In zones lying (not too far) simulate the tides are unlikely to be reliable. upstream of important sources, the tides therefore cause an increase of the average concentrations, otherwise the average 4.2.2. Upstream concentrations and WWTPs concentrations tend to decrease. In this particular case of the Table 3 clearly shows that from the two inputs considered in Scheldt, this means that the extent of the region influenced by this study (upstream concentrations and WWTPs), the main the high Rupel concentrations is significantly enlarged by the “source” of E. coli in the Scheldt is what comes through the presence of tides, mostly upstream but also downstream. upstream boundaries. This is probably due to the fact that Conversely, the most upstream section of the Scheldt is mainly influenced by what comes from further upstream, and only to (i) a huge amount of bacteria enter the model domain thr- a lesser extent by the tide. In this part, the tides rather have the ough the Zenne boundary, caused by the large volumes of water research 45 (2011) 2724e2738 2735

4000 medians IQRs

2000 −1

0 E.coli (100 ml) −2000 Uitbergen a Temse −4000 0 Southern ↑ 10 20 30 Dender ↑ 40 50 Durme ↑ 60 Rupel ↑ 70 Scheldt branch km

100 medians IQRs

% 0

−50 Uitbergen b Temse −100 0 Southern ↑ 10 20 30 Dender ↑ 40 50 Durme ↑ 60 Rupel ↑ 70 Scheldt branch km

Fig. 9 e Difference between simulation with tides and simulation without tides. (a) Absolute difference and (b) relative difference between medians (full line) and interquartile ranges (IQRs, dashed line). Positive values mean that the simulation with tides is associated with higher median or IQR. The location of tributaries joining the Scheldt and the two monitoring stations Temse and Uitbergen are also indicated.

waste water discharged in the Brussels area (upstream of implying that bacteria need to cross a significant water the model boundary) in the relatively small river Zenne. depth before they actually disappear by settling. The (local) These massive concentrations propagate through the relative importance of mortality versus sedimentation can ¼ Rupel into the Scheldt, where they overwhelm the effect be expressed as q kmortH/vsed,withH the water height. In of local WWTPs. the freshwater (1D) part of the Scheldt and during the study (ii) the largest WWTPs in the Scheldt (the part under tidal period (26 March 2007e15 June 2008) this ratio ranges influence) have a limited effect. Most of them are located between 2 and 35, with a median value of 9. In other words, in the Antwerp area, where they either discharge in disappearance by mortality is always faster than by sedi- canals or in the Antwerp harbour, avoiding a direct effect mentation. For the Seine watershed, it was already found on the Scheldt. The few large WWTPs that discharge that the relative importance of settling versus mortality in directly in the Scheldt (e.g. Antwerpen-Zuid and Aartse- the total disappearance rate decreases with increasing laar), do so in the downstream part of the river (down- hydrological order of the stream (Servais et al., 2009). For stream of the Rupel connection) where water discharges small streams, settling was the dominant cause of E. coli are much higher and therefore their impact is immedi- disappearance, while its importance became negligible in ately reduced by dilution. the largest rivers of the watershed. Nevertheless, we must keep in mind that the settling process was modelled by Although Table 3 only focuses on Temse and Uitbergen, the means of a very simple first order parameterisation, while concentrations are reduced in the whole domain when a more accurate representation would include an explicit the boundary concentrations are set to zero (not shown). The model of suspended matter (including resuspension). It was effect of the WWTPs is then more visible but remains only already discussed that such a representation is expected to very local, suggesting an efficient mixing/dilution. improve the model performance in the estuarine MTZ. However, it is not obvious whether it will significantly 4.2.3. Disappearance processes influence the results in the riverine part. Based on the E. coli Finally, we tested the impact of taking out either of the two concentrations measured in the bottom sediments and the considered disappearance processes: mortality and settling. concentrations of suspended matter, Ouattara et al., (2011) Table 3 shows that sedimentation has a negligible effect, but estimated the potential contribution of sediment resus- mortality certainly not. In other words, it is the mortality pension to the E. coli concentration in the water column. process which is primarily responsible for the decrease in Sediment resuspension contributed significantly to the concentrations following the input by a WWTP or tributary water contamination only at two sites in the Scheldt (Fig. 3). The negligible importance of the settling process on watershed. These results suggest that resuspension can the overall disappearance rate is probably due to the fact have important but localised impacts in the rivers. Model- that the rivers considered in this study are relatively deep, ling these effect will be a challenge for the future. 2736 water research 45 (2011) 2724e2738

0.02 a 0.015 0.01

0.01 0.005 concentration 0 0.005 0 0.1 0.2 0.3 0.4 tracer concentration days 0 0 2 4 6 8 10 days

60 b

20 with tide without tide

km from Ghent (1D model) 0 0 2 4 6 8 10 days

10 c 8

6 km 4

0 0 2 4 6 8 10 days

Fig. 10 e Results of simulation in which a patch of passive tracer was released at Uitbergen. No sources or decay processes were considered. In a first simulation, the tracer was transported by the “full” hydrodynamics (tides D river flow, thick lines) and in a second simulation only the river flow was considered (thin lines). Results shown: (a) evolution of maximal concentration (arbitrary units), with inset showing zoom on first hours; (b) position of maximum; (c) measure for the width of the patch.

kept relatively simple, motivated by analogous studies (e.g. 5. Summary and conclusions disregard of diffuse sources) and by lack of data (constant WWTP discharge, boundary concentrations, single pool of The current study aimed at providing some insight into the bacteria). Nevertheless, the model simulations were capable (observed) E. coli concentration in the tidal Scheldt River and of reproducing the long-term median and range of E. coli Estuary. At a few locations along the tidal Scheldt long-term concentrations in the Scheldt. The main deficiency of the monitorings (>1 year) have been performed, and the resulting model is its inability to accurately simulate the decrease in (monthly) measurements exhibited a remarkable variability, concentration in the MTZ e which is most probably due to the which could not readily be explained. Although measure- lack of sediment-related dynamics for E. coli. ments are available only at a monthly interval, we hypoth- This is not the first E. coli model resolving the tide, but esised that the short term physical processes (tide and previous studies did not investigate the long-term effect of upstream discharge) could be major drivers. To verify this this forcing. Kashefipour et al. (2002) focus on single days, hypothesis, the SLIM-EC model was built, in order to simulate Garcia-Armisen et al. (2006) only study the concentration the spatio-temporal distribution of E. coli concentrations, profile after a 28 days simulation with constant upstream including these high-resolution physical forcings, in addition discharge. On the other hand, we must admit that the current to specific E. coli sources (WWTPs, boundary concentrations) model is not fit for “point predictions” at a precise time and and processes (mortality and settling). The E. coli dynamics are location. Still, the model has proven accurate in predicting water research 45 (2011) 2724e2738 2737

long-term median and range, making it a potentially inter- researcher with the Research Foundation Flanders (FWO), and esting tool for long-term risk assessment studies. Indeed, for with the Belgian National Fund for Scientific Research risk studies, understanding of the median behaviour is not (FRS-FNRS). Eric Deleersnijder is a Research associate with the sufficient; it is crucial to have some insight into the variability Belgian National Fund for Scientific Research (FRS-FNRS). The and the processes driving it. research was conducted within the framework of the Inter- Comparing the reference simulation to reduced model university Attraction Pole TIMOTHY (IAP VI.13), funded by the setups, a deeper understanding of the controlling processes Belgian Science Policy (BELSPO). SLIM is developed under the was possible: auspices of the programme ARC 04/09-316 and ARC 10/15-028 (Communaute´ française de Belgique). (1) The tide, the concentrations coming from upstream and the mortality process are the main factors causing the references observed E. coli concentrations and variability. (2) The tide is crucial to find correct median and range of concentrations. However, its effect is complex: it can either increase or decrease the local (median) concentra- Baeyens, W., van Eck, B., Lambert, C., Wollast, R., Goeyens, L., tions (depending on the location of the closest sources) 1998. General description of the Scheldt estuary. Hydrobiologia 366, 1e14. and increase or decrease the local variability. Barcina, I., Lebaron, P., VivesRego, J., 1997. Survival of (3) The impact of the WWTPs inside the model domain are allochthonous bacteria in aquatic systems: A biological minor, suggesting that investment in these WWTPs may approach. Fems Microbiology Ecology 23, 1e9. not be the most efficient management action to improve Chen, M.S., Wartel, S., Van Eck, B., Van Maldegem, D., 2005. the water quality in terms of fecal contamination. At the Suspended matter in the Scheldt estuary. Hydrobiologia 540, e opposite, improving wastewater treatment in some 79 104. Collins, R., Rutherford, K., 2004. Modelling bacterial water quality WWTPs located upstream of the studied domain (especially in streams draining pastoral land. Water Research 38, in the Brussels area) would be important from a water 700e712. quality point of view. Craig, D.L., Fallowfield, H.J., Cromar, N.J., 2004. Use of macrocosms to determine persistence of Escherichia coil in These results point towards a few directions for future recreational coastal water and sediment and validation with developments: in situ measurements. Journal of Applied Microbiology 96, 922e930. Davies, C.M., Bavor, H.J., 2000. The fate of stormwater- (1) Model improvements: associated bacteria in constructed wetland and water pollution control pond systems. Journal of Applied a. A better model representation of the estuarine decrease Microbiology 89, 349e360. in E. coli concentrations may be achieved by complexifying the Davies, C.M., Long, J.A.H., Donald, M., Ashbolt, N.J., 1995. Survival E. coli module by including a direct link with sediment of fecal microorganisms in marine and fresh-water dynamics. sediments. Applied and Environmental Microbiology 61, e b. Include further variability in the forcings, especially the 1888 1896. de Brauwere, A., De Ridder, F., Gourgue, O., Lambrechts, J., boundary concentrations. Including varying WWTP discharges Comblen, R., Pintelon, R., Passerat, J., Servais, P., Elskens, M., does not seem relevant, due to the small impact of these Baeyens, W., Ka¨rna¨, T., de Brye, B., Deleersnijder, E., 2009. sources. However, a more accurate representation of what Design of a sampling strategy to optimally calibrate a reactive enters from upstream could be achieved by extending the transport model: Exploring the potential for Escherichia coli in model to the more upstream (non-tidal) river sections, espe- the Scheldt Estuary. Environmental Modelling & Software 24, e cially the Zenne section crossing Brussels, as this appears to be 969 981. de Brauwere, A., Deleersnijder, E., 2010. Assessing the a major source of contamination. parameterisation of the settling flux in a depth-integrated model of the fate of decaying and sinking particles, with (2) Additional data. Indeed, the above-mentioned model application to fecal bacteria in the Scheldt Estuary. improvement are only possible if additional measure- Environmental Fluid Mechanics 10, 157e175. ments are made/become available. But also for the vali- de Brye, B., de Brauwere, A., Gourgue, O., Ka¨rna¨, T., Lambrechts, J., dation of the model additional data are necessary. Visually Comblen, R., Deleersnijder, E., 2010. A finite-element, multi- it is clear that data (timeseries) are lacking in the estuary, scale model of the Scheldt tributaries, River, Estuary and ROFI. Coastal Engineering 57, 850e863. but also in the riverine part additional monitoring stations Edberg, S.C., Rice, E.W., Karlin, R.J., Allen, M.J., 2000. Escherichia would be useful. The model may be a useful guide to coli: the best biological drinking water indicator for public determine the optimal position and/or timing of future health protection. Journal of Applied Microbiology 88, samples (e.g. de Brauwere et al., 2009). 106Se116S. EEA, 2004. Impacts of Europe’s Changing Climate. An Indicator- based Assessment. Report n2/2004. Office for Official Acknowledgements Publications of the EC, Luxembourg. EU, 2000. Directive 2000/60/EC of the European Parliament and the Council of 23 October 2000-Establishing a framework for The authors wish to thank the Vlaamse Milieumaatschappij Community action in the field of water policy. 72p. for providing data on fecal coliforms in the Scheldt. Anouk de EU, 2006. Directive 2006/7/EC of the European Parliament and of Brauwere performed this study while she was a postdoctoral the COuncil of 15 February 2006 concerning the management 2738 water research 45 (2011) 2724e2738

of bathing water quality. Official Journal of the European near-shore region of lake michigan. Environmental Science & Union 64, 37e51. Technology 40, 5022e5028. Garcia-Armisen, T., Prats, J., Servais, P., 2007. Comparison of Muylaert, K., Sabbe, K., 1999. Spring phytoplankton assemblages culturable fecal coliforms and Escherichia coli enumeration in in and around the maximum turbidity zone of the estuaries of freshwaters. Canadian Journal of Microbiology 53, 798e801. the Elbe (Germany), the Schelde (Belgium/The Netherlands) Garcia-Armisen, T., Servais, P., 2007. Respective contributions of and the Gironde (France). Journal of Marine Systems 22, point and non-point sources of E. coli and enterococci in 133e149. a large urbanized watershed (the Seine river, France). Journal Okubo, A., 1971. Oceanic diffusion diagrams. Deep-sea Reserach of Environmental Management 82, 512e518. 18, 789e802. Garcia-Armisen, T., Servais, P., 2008. Partitioning and fate of Ouattara, N.K., Passerat, J., Servais, P. 2011. Faecal contamination particle-associated E. coli in river waters. Water Environment of water and sediment in the rivers of the Scheldt drainage Research 81, 21e28. network. Environmental Monitoring and Assessment. doi:10. Garcia-Armisen, T., Thouvenin, B., Servais, P., 2006. Modelling 1007/s10661-011-1918-9. faecal coliforms dynamics in the Seine estuary, France. Water Prats, J., Garcia-Armisen, T., Larrea, J., Servais, P., 2008. Science and Technology 54, 177e184. Comparison of culture-based methods to enumerate George, I., Crop, P., Servais, P., 2002. Fecal coliform removal in Escherichia coli in tropical and temperate freshwaters. Letters wastewater treatment plants studied by plate counts and in Applied Microbiology 46, 243e246. enzymatic methods. Water Research 36, 2607e2617. Rozen, Y., Belkin, S., 2001. Survival of enteric bacteria in seawater. Geuzaine, C., Remacle, J.F., 2009. Gmsh: A 3-D finite element mesh Fems Microbiology Reviews 25, 513e529. generator with built-in pre- and post-processing facilities. Sanders, B.F., Arega, F., Sutula, M., 2005. Modeling the dry- International Journal for Numerical Methods in Engineering weather tidal cycling of fecal indicator bacteria in surface 79, 1309e1331. waters of an intertidal wetland. Water Research 39, Havelaar, A., Blummenthal, U.J., Strauss, M., Kay, D., Bartram, J., 3394e3408. 2001. Guidelines the current position. In: Fewtrell, L., ServaisP, Billen, G., Garcia-Armisen, T., George, I., Bartram, J. (Eds.), Water Quality: Guidelines, Standards and Goncalves, A., Thibert, S., 2009. La contamination Health. In World Health Organization Water Series. IWA microbienne du bassin de la Seine. Programme Publishing, London. Interdisciplinaire de Recherche sur l’Environnement de la Kashefipour, S.M., Lin, B., Harris, E., Falconer, R.A., 2002. Hydro- Seine. PIREN-Seine, ISBN 978-2-918251-07-1. environmental modelling for bathing water compliance of an Servais, P., Billen, G., Goncalves, A., Garcia-Armisen, T., 2007a. estuarine basin. Water Research 36, 1854e1868. Modelling microbiological water quality in the Seine river Kay, D., Bartram, J., Pru¨ ss, A., Ashbolt, N., Wyer, M.D., Fleisher, J.M., drainage network: past, present and future situations. Fewtrell, L., Rogers, A., Rees, G., 2004. Derivation of numerical Hydrology and Earth System Sciences 11, 1581e1592. values for the World Health Organization guidelines for Servais, P., Garcia-Armisen, T., George, I., Billen, G., 2007b. Fecal recreational waters. Water Research 38, 1236e1304. bacteria in the rivers of the Seine drainage network (France): Lambrechts, J., Comblen, R., Legat, V., Geuzaine, C., Remacle, J.F., sources, fate and modelling. Science of the Total Environment 2008. Multiscale mesh generation on the sphere. Ocean 375, 152e167. Dynamics 58, 461e473. Thupaki, P., Phanikumar, M.S., Beletsky, D., Schwab, D.J., Liu, L., Phanikumar, M.S., Molloy, S.L., Whitman, R.L., Shively, D. Nevers, M.B., Whitman, R.L., 2010. Budget analysis of A., Nevers, M.B., Schwab, D.J., Rose, J.B., 2006. Modeling the Escherichia coli at a southern lake michigan beach. transport and inactivation of E. coli and enterococci in the Environmental Science & Technology 44, 1010e1016.